Migrate PVC from one storage class to another

There could be a requirement where you might need to migrate the PVC and its data from one storage class (storage back-end) to another. This could be because of migration from one storage solution to another due to cost and performance factor.

There are two ways we can migrate the PVC from one storage class to another.

1. Cloning PVC (volume) from one storage class to another using csi-driver
2. Migrate PVC using korb tool

Migrate PVC from one storage class to another

In this article we will see both of the ways one by one.

Contents

1. Why it works on my machine ?
2. Clone PVC using CSI-driver
3. Migrate PVC using Korb tool

Why it works on my machine?

I am using:

• Kubernetes version 1.28.7
• One control plane and two worker nodes on Ubuntu 22.04
• Containerd CRI 1.6.28
• csi-driver-nfs v4.5.0 as csi-driver for nfs with external snapshotter enabled
• korb version 2.2.0

Clone PVC using csi-driver

PVC cloning feature is only supported for csi-based PVCs using CSI drivers.

A Clone is a duplicate of an existing Kubernetes Volume that can be consumed as any standard Volume would be. The only difference is that upon provisioning, rather than creating a new empty Volume, the back end device creates an exact duplicate of the specified Volume.

Below requirements must be met to clone a PVC:

1. CSI driver of storage provisioner must support cloning feature. Please check with your csi-driver provider to know if PVC cloning feature is supported.
2. Cloning can only be performed between same volumemode storage. For e.g. source volume mode is filesystem, cloned PVC must also have filesystem volume mode.
3. Make sure csi external snapshotter is deployed on your cluster. Check with your csi-driver provider to deploy it on cluster.
4. The source PVC must be bound to a PV.

For this demo, I have two storage classes source nfs-csiand target new-nfs-csi. Both are backed by NFS servers running on two different servers.

Let’s start our demo:

1. Create storage class for target storage provisioner. I have target storage class with name new-nfs-sc.

Create storage class for target storage

2. I already have a PVC deployed on my cluster with source storage class and mounted to a pod.

Source PVC

3. Let me copy a 1G of file inside PVC using kubectl cp command for testing.

I have nginx pod on which source PVC is mounted at /var/log/nginx path.

Copy test file inside PVC

4. Create PVC clone using below yaml

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: nginx-logs-clone  # <------------- Cloned PVC name
spec:
  storageClassName: new-nfs-sc  # <------- Target storage class
  dataSource:  # <------------------------ This field provides the source of PVC
    name: nginx-logs # <------------------ Source PVC name
    kind: PersistentVolumeClaim
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 3Gi

As you can see in above YAML, there is a dataSource field where we can mention source PVC details. Kubernetes and CSI-driver uses that field to identify the source PVC which need to be cloned. storageClassName field has storage class name of target storage class.

The cloned PVC size must not be less than size of source PVC.

Deploy cloned PVC

Once we apply above YAML file, We will see that a new PVC is created. It will create a PV in target storage class and copy the content of source PV to cloned PV.

5. Now lets check the content of cloned PV.

A. To check the content, We will mount the cloned PV as well to same pod by editing the nginx deployment.

Edit deployment to mount cloned PVC

Mount cloned volume to pod

B. Compare the content between source PVC and cloned PVC

Content of source and cloned volume

MD5 checksum of 1G file in source and cloned volume

Now that you see PVC can be cloned using csi-driver. There are some limitations on this. Some of the limitations are:

1. This way is only possible with csi-backed PVCs
2. CSI-driver provided by storage provisioner must support volume cloning feature
3. We need exxternal snapshotter to be deployed on cluster

To know more about volume cloning, Follow Kubernetes document — https://kubernetes.io/docs/concepts/storage/volume-pvc-datasource/

Migrate PVC using korb tool

Korb is a open source tool which is used to migrate a PVC from one storage class to another. This works with any type of storage class and need not be csi-backed.

It internally uses rsync and copies the content from PVC on one storage class to another and then deletes the source PVC.

To use Korb we must download Korb tool on jumpbox or VM from where we can connect cluster.

curl -LO https://github.com/BeryJu/korb/releases/download/v2.1.3/korb_2.1.3_linux_amd64.tar.gz
tar -xvzf korb_2.1.3_linux_amd64.tar.gz
sudo mv korb /usr/local/bin/korb

Currently I have PVC in nfs-csi storage class. I willl migrate it to new-nfs-csi storage class.

Before cloning a PVC, Pods using the PVC must be scaled down.

Scale down nginx pod using source PVC

Now, To clone the PVC, use below command

korb --new-pvc-storage-class=<name-of-target-storage-class> --strategy=copy-twice-name <source-pvc-name>

Above command will create the PVC with same name as source PVC but in different storage class with same content. It may take time to copy data from source to migrated PVC depending on data size.

If you want to create PVC with different name, --new-pvc-name flag can be used with new pvc name argument.

Migrate PVC to another storage class using Korb

Lets check the new PVC. Now the PVC is created with same name but in another storage class new-nfs-sc.

PVC migrated to another storage class

Now let’s scale up nginx pod and check the content inside PVC. Since PVC name is still same, I don’t need to edit deployment to mount new PVC.

As you can see the content is retained in new PVC with new storage class.

You can read more about Korb tool here — https://github.com/BeryJu/korb

I hope this article must have given you idea about migration of PVCs from one storage class to another.

You can support my work by buying me a cup of coffee on https://www.buymeacoffee.com/linuxshots

Thank You!

Navratan Lal Gupta

Linux Shots